February 25, 2025 • 47 mins
Today on Data Nation, we are speaking with Kerry Emanuel and Kelvin Droegemeier to better understand extreme weather prevention and technology. Together, we’ll explore extreme weather displacement, changing weather conditions, and the legislation behind natural disaster procedures and measures.
With Kerry Emanuel’s expertise in meteorology and extreme weather prediction as well as Kelvin Droegemeier’s legislation work as the Former Director of the Office of Science and Technology Policy, today’s episode will provide insightful and credible conversation on this topic.
With millions of Americans at risk from climate change, it is crucial to understand where laws and technology are shifting in the world of extreme weather events.
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:06):
[music]
Today on Data Nation, we're speaking with Kerry Emanuel, co-founder of the Lorenz Center at MIT,
and Kelvin Droegemeier, former director of the Office of Science and Technology Policy,
a position otherwise known as Chief Science Advisor to the President of the United States,
in order to dive deeper into natural disasters, climate change, and extreme weather prediction
(00:29):
technology. I’m Liberty Vittert, professor at Washington University in St. Louis, and my co-host
is Munther Dahleh, William A. Coolidge professor in electrical engineering and computer science at
MIT. In 2024, Americans all over the country faced loss and damage from natural disasters
and extreme weather events, from LA wildfires to hurricanes Helene and Milton in the Southeast.
(00:54):
The NOAA’s National Center for Environmental Information estimated that extreme weather
events in America caused almost $200 billion worth of damages in 2024 alone. One year prior,
the US Census estimated that 2.5 million Americans were displaced from their homes
by weather-related disasters. With millions of Americans at risk of losing their homes
(01:19):
and communities due to weather conditions, climate experts and scientists have been
working relentlessly to develop premeditated and predictive weather technologies to limit the
damage from natural disasters. With 2024 recorded as the warmest year on record according to NASA,
climate change and natural disasters will continue to be a problem for everyone on Earth. With the
(01:44):
challenges of global warming and the pressures to prevent widespread damage to Americans. We
are speaking with Kelvin Droegemeier and Kerry Emanuel to better understand what legislative
and scientific changes can be made to combat future issues from natural disasters in America.
Munther (02:03):
You know, I think the first question that came to my mind is about the fires in LA,
150,000 people displaced. Did we mispredict what happened that got us to that point and
is this about sort of advancing the knowledge and the science behind the
modeling? What's missing and put us in the right track on this particular issue.
Kerry (02:24):
I can say that there's a whole kind of cottage industry in
what's called extreme event attribution, to what extent can you take anything like
a big hurricane or a wildfire of the kind that affected Los Angeles and say anything
scientifically credible about the extent to which climate change has had an effect on that,
(02:48):
and I don't know what exactly the answer is for that particular event. Although climate
models generally do predict that conditions for wildfires in the West writ large will becoming
more favorable for those fires, but that's not to say that particular fire. I want to say that
(03:09):
that when I look at problems like that I look at it in the context of a non-climate trend,
which is a strong demographic trend. People all over the United States to be moving from
less risky to more risky areas. It’s a trend that arguably at least partially takes place
because we inadvertently subsidize that trend, and that's something that you might say that
(03:35):
climate change is a threat multiplier of that, as people move into riskier places.
That's something that we could do something about from a policy perspective, and probably
should irrespective of whether climate change is becoming more dangerous, but certainly it is.
Kelvin (03:53):
I’d like to build on what Kerry said about attribution science. It is a science of
attributing a particular event or class of events to other forcing phenomena like climate change and
other things like that, and it's gotten quite good in in the past several years, it's advanced rather
rapidly and now there's a technique called pseudo global warming where when we run our
(04:13):
climate models Earth system models for very long periods of time we start them with sort of a some
reference climate state and we run them to kind of an equilibrium and then we start turning on
various forces and things like that, but to your point about the the wildfires on a particular day
or particular part of the year how do we do that? So one of the ways to do that is take a given
(04:34):
event that has happened and then look at what the climate would have been like prior to that event
and then take that delta and resimulate that event and really not be able to say just a class of
events can be attributed to some other phenomena like climate change but a specific event. Now,
as Kerry said you it's very difficult to do that in a lot of cases. Wildfires, of course,
(04:55):
have multiple dimensions to them and it's very much like my own experience in working a lot of
aircraft accidents, commercial accidents, it's not typically one causal factor it's typically
multiple things line up and they line up a particular way to cause something to happen. So it
is a difficult thing to tease out, but I will say that I think we are getting better at doing that
and so I think from a policy perspective the real key is to understand what things we really can
(05:19):
influence on the short term and the long term, and then take actions in a policy sense and other ways
to look at effectuating some type of of change to prevent those things from happening in the future.
Munther (05:29):
So, in general, predicting these extreme events is very difficult, and then there's the
question of having the right emergency response system in place. It’s not so much about having the
accuracy, and for example when we think about the failure of the outages of the power grid,
again those are the kind of things where it's very difficult to predict the cascaded failures
(05:52):
that can occur over there, but there are some simple principles - and we don't do that very
well - but there are simple abstracted models that can give you an idea of when
to start mitigating some of that risk. Is this analogy extendable to situations like
fires or hurricanes or - in the extreme case, not in the more predictable cases?
Kelvin (06:16):
Well, I think pre-positioning assets and anticipating an event is extremely important,
and one of the things that that is being worked on now around the world is to take
our Earth system models which operate relatively coarse meshes in terms of the fidelity that they
have to resolve fine scale features and bring them down essentially to the scale
(06:37):
of what we use now in numerical weather forecasting, over say a two or three week
period. When you're running an Earth system model for multiple years or multiple decades,
we don't have the computational power to run at that fine fidelity which provides the much more
localized decision-making information that people would need to anticipate events and prepare for
them. So it's a difference between very strategic positioning and very tactical positioning. So one
(07:02):
of the the things that is is hoped for for the future, and of course artificial intelligence
influencing this as well, is to really be able to provide that very localized fine-scale
information over multiple decades actually for things like agriculture, for transportation,
for the kinds of things we're talking about here. We’re not there yet, and we've used the concept
(07:22):
of downscaling where you actually provide a much finer mesh of the model over some area,
and that is really excellent thing to do but there are some limitations with that. So there are about
over a dozen groups around the world now that are working on these what are called kilometer-scale
Earth system models, and moving from what we have been using historically to that type of
fine-scale is a tremendous challenge, it takes a long time to do that. Plus, we actually then have
(07:48):
to figure out - how do we initialize those kinds of models, perhaps even using real observations,
which is really not something that is done now. So it's a very exciting area of science that is,
I think, helping position us to answer exact the kind of question that you posed. If you
talk to decision makers, you talk to people in the reinsurance industry, the insurance industry,
people who make these more localized decisions are - they will tell you ‘we're not really
(08:09):
getting the information that we need’ and so I think listening to those stakeholders and having
them drive the agenda to where we make sure that we're paying attention to what their needs are,
as well as understanding the science of the earth system which is extremely important.
We have to do both and they're very, very much congruent with one another.
Liberty (08:26):
Just to sort of follow on, Kerry, from something you mentioned about people moving to
sort of high-risk areas. I don't remember where I read it but you know it was like,
Miami's built on a swamp, there's going to be more damage when it says there's ten times the amount
of money, more damage. it's like well that's that's how it's going to work when you have a
(08:46):
lot more people living there, and there was that recent study that found that human-caused climate
change made Atlantic hurricanes, I think it was 18 miles per hour stronger in the last six years
and you had had said that the study makes and is similar to your predictions from almost 40 years
(09:06):
ago. So could you give us sort of a little bit more in-depth discussion on how climate change
is worsening natural disasters, or is it both, is it that it's worsening things but we also
have people living in Miami on a swamp, and what are what are your sort of future predictions for
the strength of natural disasters and is there hope, really for dampening effects in the future?
Kerry (09:31):
Well it's a long and interesting string of questions.
Liberty (09:34):
Sorry, it was my stream of consciousness here.
Kerry (09:37):
I think we have to take this hazard by hazard. I've worked a lot, as you pointed out,
on hurricanes, and the community gets together, the community of people who research hurricanes
and climate get together periodically and compare notes. I think there's pretty strong confidence
that hurricanes are becoming more intense. We’re beginning to see it in the satellite data.
(09:59):
They not necessarily becoming more frequent except in the Atlantic region and that is for
reasons that may not have a lot to do with global warming, and very importantly they rain more, and
what many people don't realize is that the major source of loss of life in hurricanes is not wind,
it's water, either from torrential rain-induced flooding or I should say and/or from storm surges.
(10:27):
So they're getting worse but that's compounding the other phenomenon you mentioned, and this could
be said of some other hazards as well, of people moving into places not not just living there,
people have lived in Miami for a while, but actually actively moving from less risky and
more risky places and this is very much on my mind because it's very hard to reverse that. Why
(10:51):
is that happening? Well there are all kinds of ways we subsidize people to live in places they
shouldn’t, and that is to protect them from the cost of that. It’s complicated and it has to do
with the way insurance is set up and regulated, and it sounds humane on short time scales but
it's catastrophic on long time scales. So, for example, all of the water damage which
(11:15):
is usually a lot of it from natural hazards like floods and hurricanes is insured by the federal
government through the national flood insurance program, that's been true since 1968, and because
it is a government enterprise it's captive to politics and people usually well-connected,
(11:37):
relatively well-to-do people living in dangerous places agitate for their rates to be kept low, and
indeed they are. So, for example, where I live on the coast of Maine it's not exactly Palm Beach but
it's wealthier than inland, we get in my county four times as much money payouts from the National
(11:59):
flood insurance program as we collectively put into it through premiums. In Western Kansas,
that ratio is reversed. People are putting four times as much premiums in as they're getting back,
and that's unfair I would say in the first place but worse than that it's encouraging
this migration. There are all kinds of other ways this happens. But on the good side of it,
(12:23):
and to add on to something that Kelvin said we actually have gotten collectively much much better
at warning people in the face of an immediate disaster, at least for some kinds of hazards like
hurricanes and floods, and the death toll from these things has gone way down in spite of the
fact or at the same time the damage is going way up, and we can do better and we should do better
(12:48):
and I don't think we're anywhere near that level of competence for wildfires but we're headed that
way. So it seems like a tragedy that we're able to save lives but also we're suffering a lot more
damage at the same time, that's the general view of the natural hazard landscape from where I sit.
Kelvin (13:10):
If I could just build on something that Kerry said and take it even to a
broader perspective, this is something that the research community an the atmospheric sciences and
meteorology and the climate sciences community began to really recognize probably twenty or
so years ago and that is that all the physical science and technology in the world isn't going
to prevent people from dying, isn’t going to prevent some of the catastrophic loss, and that
(13:32):
really we're dealing with the coupled human and natural systems. So the social-behavioral
sciences and the economic science dimensions of what we're talking about here, in particular
hurricanes, is extremely important. So now we're studying these problems really in their totality,
bringing in policy dimensions as well from the very beginning rather than just be at the end
of where it's like ‘okay we now understand the physical science of this phenomena,
(13:54):
now what do we do with that understanding?’ We have those folks at the table from the very
beginning, because this truly is a coupled, very entangled and interactive kind of system and so
to achieve the things that Kerry talked about is really, it means to understand human behavior, how
we communicate risk, how humans respond to risk and decision-making under stress, what evacuation
(14:14):
pathways exist and so on and so forth, and for example there's a high risk or some type of a
risk or a threat of say severe thunderstorms we do a great disservice by using that term because
risk depends on your own particular situation - where you are, where you live, is English
your first language, your risk is determined by a lot of factors compared to what the atmosphere is
(14:35):
going to do and so this is an area of scholarly work that is really getting a lot of attention
now to convey risk in a way that is sort of customized to your own particular situation at
the moment or moments preceding the event actually happening, which I think is going to add to what
Kerry said in terms of increasing the ability to prevent the loss of life and hopefully protecting
property and reducing loss. So I just thought it would be important to kind of mention that.
Munther (15:00):
I think that's really interesting, and at some level it's comforting to know
that less people are dying even though the damage in terms of cost is very high,
but the point is tailoring the knowledge of the situation to the particular person
I think it's really a fascinating idea. So to segue a little bit, people are scared,
(15:22):
a lot of people have these conversations about ‘what if the unexpected happens in my region?’ you
know it's like maybe earthquakes in places that don’t have earthquakes or a sudden flooding in a
different area and so forth. So we have that one aspect, there's all the science with the
science community and what they know and there's a mistrust, of course, between the communities that
(15:45):
we talk about and then there's the government and Kelvin you have been the head of the White
House Office of Science and Technology Policy and so there's the aspect of that which is
regulated and mandated by the government. How is that working out all together? All these
different components and what is the government actually doing to help in these situations?
Kelvin (16:05):
Right, no that's a really great question and ultimately the federal government
has responsibility to protect its citizens, to ensure economic vitality of the nation
and things like that. What a lot of times folks don't understand is that when you develop policy,
my role at OSTP was to make sure it was at the table, but policy actually is not just
(16:25):
what science says is the case, and of course we learned and saw in the pandemic the public
really got I think a peek behind the door to see how science really works, it's not here's
an immutable answer to this particular thing, do I wear a mask or don't I? This study said one thing,
this study contradicted and that's that's how science has always worked, but I don't think the
public generally understood that, they just wanted to know do I or don't I wear a mask, just tell me,
(16:47):
and so when you think about the policies that get formulated especially in situations like that,
science is at the table but there's so many other factors that come into play. There's economics,
there's certainly politics there's national security, all these kinds of things and you
see the policy come out the other side and you're like ‘well wait a second, that seems incongruent
with what we think we understand about about the natural world, about hurricanes or whatever’ so
(17:09):
it's a very complicated process and sometimes it doesn't turn out the way you would perhaps
hope it is because one part of that policy forcing dominates another, and so really all that we can
do as a scholarly research community is make sure that that we convey the facts as we best
understand them, we do it in such a way that that other folks can really take that up and use it and
(17:30):
then translate it into meaningful policy that is actionable, that has intended outcomes, and
doesn't have any serious unintended consequences which sometimes is is quite difficult, but I think
that the government works fairly well most of the time, but when you really start to understand how
policy happens and how the sausage is made then it begins to make sense of why some of
the things are the way they are and they're not the way you would think they should be.
Liberty (17:52):
Thinking about the future of all of this, Kerry, I saw a research article that you did sort
of proposing TC-GEN and the article described from my understanding TC-GEN is a machine learning
based framework that basically uses historical data sets and climate models to reproduce these
(18:14):
really high-resolution weather conditions, and I wanted to know in terms of the concepts of the
future, can this past data, especially given as you said with high-risk areas and how things are
changing so quickly and people are moving to these places, can it really help sort of reproduce these
conditions now, and do you see this technology as really helping disaster preparedness for regions?
Kerry (18:40):
In spite of the fact I was a co-author on that paper, I actually prefer the kind of
approach that Kelvin was talking about earlier, it's called downscaling. Let me try to explain
why I do that . So one of the problems we have is if you look at places that have 100 years of
good records of weather and also of damage that weather causes, you discover that over a very
(19:08):
long period of time if you accumulate damage from that hazard most of the damage comes from events
whose frequency is roughly once in a 100 years. The very frequent events we’re well adapted to
so they don't cause much damage and if they're too rare, well they're so rare that they don't
contribute in the long run. Well 100 years, to get a really robust estimate of how much damage a 100
(19:34):
year event would occur you need about a thousand years of records. Well, we don't have that,
in fact we're lucky if we have a hundred. So even if the climate weren't changing, we have a problem
that we're not able through the historical record to really see many of the events that in the long
run, like the Los Angeles fires, the rare events that cause a lot of damage, and so I've been on a
(19:58):
campaign to try to use physics through models to evaluate today's risk. That whole problem is
made much worse by climate change and I think it's true to say for many phenomenon that even
if we had a thousand years of beautiful records, arguably they're not relevant to today because of
(20:19):
the climate change that's already happened, and insurance companies are very rapidly waking up
to this fact. All of their risk models, all the decisions they make and a lot of cities and towns
and states are making are based on historical data which is no longer relevant to today. So this
(20:39):
is part of the reason that we're seeing so much damage is that we're going outside this envelope,
but as great as artificial intelligence is it usually fails when you try to push it outside the
set of data that you use to train it, and that's automatically true with climate change. You just
(21:00):
don't have data pertinent to today's climate, you don't have nearly enough of it to do that.
So I think downscaling is a way out of that and I've tried very hard to do that in the case of
of hurricanes, and we are seeing that in many places including the Carolinas, the risk today
is much higher than it was 20 or 30 years ago and yet it's not being reflected in insurance rates,
(21:25):
it's not being reflected in construction practices and building codes. So that's where
a lot of this damage is coming from, we're just not adapting fast enough to the changing climate.
Munther (21:38):
But if I jump in and say that - trying to understand what you meant by bringing in physics,
because in the world of Engineering in general what you said is absolutely true
that is basing models on pure historical data can actually miss the future. But I
think one important aspect of this is modeling all the components and
(21:59):
all the pieces and understanding the the relationships between them right,
and so the question is if temperature and heat and other factors also affect the global phenomenon,
why aren't the models developed to take into consideration all these inputs and then simulate
and learn potentially the underlying physics which has like been demonstrated in other fields.
Kerry (22:24):
Well, there's a lot packed into your very interesting question. So one of the problems is
the word model. It means very different things to different people, even to different sets of
scientists. So a numerical weather prediction model of the kind that Kelvin talked about a
few minutes ago is a very, very different piece of machinery than an economic model. That is,
(22:49):
in the first case we actually do know the laws of physics, but we can't afford to integrate them at
high enough resolution to capture even phenomena like hurricanes, and we have to get clever and
embed fine-scale models within these - there are ways of doing that. But the enormous advantage
(23:10):
if you can do that is the laws of physics are invariant, they don't change because the
climate's changing. So climate models are models that we would like to see resolve phenomena that
are on scales of several hundred kilometers or larger, they might be able to handle heatwaves. In
practice, because their resolution is limited, we have to do something called parameterization for
(23:38):
things like small-scale phenomena like convection and different models do it differently, and so one
of the things we see when we downscale different climate models is we get very different answers.
So it's not the laws of physics that are to blame for that, it's the way we encode them in models.
Kelvin (23:58):
And the one advantage that we hope to gain when we take these models that Kerry’s
talking about, and instead of having to downscale over a particular area, we actually have a uniform
resolution or nearly so across the globe, is that we then get the feedbacks across the scales. Kerry
talked about convection, these major convective systems that then feed back to the large scale,
(24:18):
the large scale influences the direction of the convective system, that cross-scale interaction
both in time and space is very important and it's not something that we really capture even
with the downscaling models, as phenomenal as they are. I've seen downscale simulations that
you’d swear you're looking at a satellite loop, it's it's just phenomenal. So the good news is
the science is really advancing, we're learning more but Kerry’s really made an important point
(24:41):
here that AI is not a singular solution. Now in weather prediction, in weather forecasting we're
seeing dramatic effects there because we do have a very, very long record and multiple records of
predictions and things like that so there's a lot more information out there to train models with,
and the interesting thing when you think about machine learning, when you train a
model it sort of knows the future. When we're using a physics-based model we initialize it
(25:04):
with what's happening up until the current time that we start the model and then it has
to forecast the future. The AI has the advantage of knowing what the future is because it's been
trained on a lot of different futures but it has the inherent limitations, as Kerry said,
that if you try to push it too far beyond the training data set you run into problems. So
there'll be a lot of folks employed in this arena for a long time because there's a lot of work to
(25:26):
be done, but a lot of it is quite promising, but there are no silver bullets, that's what Kerry
and I found in our combined 85 or 90 years of of work, there is really no such thing.
Munther (25:36):
This is really interesting I'm just wondering I'm going to push the point of data
science and machine learning on this a little bit, because what we found for example in the last five
years our ability - take the gene folding problem, okay, the gene folding problem, I
mean you're talking about a genome of over 20,000 genes. Even if each one could take a binary value,
(26:02):
the possibilities is two to the 20,000 which is enormous and the gene folding problem is
mapping that gene structure to a fold that then maps a function and what we've been able to do,
I think, with a lot of the machine learning and clever - and I would say and that was
used before - clever ideas because it's not just a brute force application is to create 30 million
(26:27):
possible new data sets just by simulation, and when we were at only 150,000 in the last 30
years with with experimentation, and with a lot of confidence that these simulated data are actually
quite accurate. So is the community embracing this kind of thinking into these geochemical models
(26:50):
and complex sort of high-dimensional models? Is that what you're seeing in your community or not?
Kelvin (26:56):
I can tell you here at the University of Illinois we have a whole major center,
a biofabrication center that does exactly that. It looks at - instead of doing Monte Carlo,
you're doing thousands and thousands of realizations - you use AI to really bring that
number down and then AI determines what the next iteration ought to be. So it's a circular kind
of a thing where you're really informing the next calculation based upon the results that you have,
(27:20):
and I'm certainly not a biologist but I've talked to these folks in molecular and cellular
biology and they're doing really extraordinary things. It’s very much more rapidly allowing us
to understand how - your protein folding and genetic structure and things like that - to
actually fabricate small molecules and that sort of thing to cure diseases,
(27:41):
to deal with inflammation and that sort of thing. So it is certainly happening to
to my degree of awareness in that area right here on my own campus just down the street.
Kerry (27:49):
Let me add briefly to that, it seems like every other article in today's journals in
my profession, in our profession is about AI and machine learning, so that's maybe not the best -
Liberty (28:02):
I feel like that’s all professions at this point.
Kerry (28:03):
It’s all professions, but I would also say that it's very good for some things like
making many, many, many replicates of weather in today's climate and as I mentioned a while ago,
just quantifying natural hazard risk in today's climate is a huge step forward,
before we actually start worrying about the future. The fact that today is very different
(28:26):
from fifty years ago is is really, really important. The other thing I would say is
kind of a no-brainer for AI is that these big models, numerical weather prediction models,
climate models, because of their need to parameterize small-scale processes,
partly because of that, they have lots and lots of knobs to turn, and so if you - that
(28:48):
you can tune physically within some plausible range - to do that manually is just impossible,
there far too many combinations like your genome example, and so AI can help us optimize those
models. I personally think that some combination of deterministic models, ensembles of such models
(29:12):
enhanced by AI will probably be the way we make the most rapid progress, but I'm just guessing.
Kelvin (29:20):
I think Kerry is spot on with that those comments, and to the last point, when you're
looking at these very sophisticated Ferrari-level Earth system models that are just extraordinary,
we're driving them like a 1950’s small car because we're constrained by the computational resources
we have. It turns out that if you really turn those things loose on the Autobahn,
(29:43):
about 95 percent of the computational resources would be taken up by what we call the dynamics
of the model, the fluid physics that Kerry talked about. Roughly five percent with these
physical parameterizations of precipitation and clouds and so on, but yet to Kerry’s earlier
point those things are extraordinarily important and that's where AI, although it doesn't really
(30:04):
improve the efficiency of the model that much, it really improves as he just said the accuracy and
ability to capture these very complicated but very important processes in a much more accurate way,
that plus producing ensembles very, very inexpensively. But as far as you're speeding
up the core component of the model not so much. GPU’s do a very good job of that but really AI
(30:28):
is is really geared toward the kinds of things that Kerry talked about.
Liberty (30:31):
So I have to ask something that I feel like we've touched on this whole way through. So,
my parents have a house in Fort Myers, well they had a house in Fort Meyers Beach,
and the hurricane two years ago, Ian, wiped it out, and so they're thinking about rebuilding
(30:52):
but trying to get insurance is almost impossible there, and Kelvin, thinking about what you said
about individual risk and the differences, for them it was fine. They were able to get out
they had another home to go to, they were fine. Some of their neighbors though, they were either
really sick so it was hard for them to get out, or they had pets that they didn't want to leave
(31:15):
and the shelters wouldn't take. That concept of individual risk is very, very interesting, but in
the same sense of how hard it is to get insurance in some places, and then Kerry as you were saying
how in other places the insurance companies haven't kept up, do the insurance companies
know something we don’t? Are they predicting - are they a lot better at predicting these
(31:38):
natural disasters coming forward than we are? I mean, because it's so hard to get it, are they
are they just doing that based upon past data or are they predicting new things that we don't know?
Kerry (31:50):
No. I can tell you with some authority because I sit on the boards of two insurance
companies and half for twenty years, it's a much different situation than most policy holders,
because - policy holders don't like insurance companies because they're being nickeled and
dimed and there are these exclusions and - it's all true, but the enemy isn't necessarily the
(32:14):
insurance company, it’s the way it's regulated. And insurance has always been regulated for the
same reason in principle that we regulate banks. We have to ensure they're solvent so when the big
storm comes they don't pack up and say ‘we're bankrupt’ and leave you high and dry. In fact,
something has been happening for a long time with insurance that goes exactly counter to that goal,
(32:39):
that is regulators started to regulate premiums. Now this sounds very esoteric,
and it doesn't sound like I'm answering your question, but I'm going to. And that has become
captive to a lot of of political pressure in every state I'm aware of, and so what what's happened
is that insurance companies don't have the degree of freedom - if they had a better model that told
(33:04):
them the risk was worse than they think it is, they can't just go ahead and adjust the premiums,
which would have the beneficial effect of sending a message to homeowners that this place is getting
riskier and maybe encouraging them to move or or take some measures to make their dwelling a
(33:24):
little bit more resilient and so forth. But we're not letting that message go through,
and so the the result is that the profit margin of insurance companies is razor thin,
razor thin. It’s a terrible business to be in and it's not being allowed to do the job it should be
doing of communicating risk through pricing. So we have this really, really bad situation,
(33:47):
and one of the problems is - you'll notice this is happening in Los Angeles it happened in Fort
Meyers, Liberty where your parents lived, it happens everywhere after disaster - mostly just
go right back and rebuild the same thing they had before, that's the instinct to do that, If it were
a rational insurance market, they couldn't do that because they couldn't afford to insure it if they
(34:11):
did. So if we did that one thing by gracefully deregulating insurance - and I'm not a free
market ideologue but in this one instance - people couldn't afford to rebuild their property in a way
that puts that property at risk. The message would come through you do that you're going to
(34:32):
have to pay through the nose for insurance, but if you do it this way, if you make your house
fire-resistant in this way and that way, you'll have a much lower premium, we're not letting that
happen. To me that is the heart of the problem, in fact it's the heart of the whole response to
climate change problem. The reason that people aren't out in the streets agitating for our
(34:53):
government and world governments to do something about climate change is we're insulating them
from the risk. We’re not letting them feel it. Your house got blown down, okay we replace it no
strings attached. This isn't working for society and if people had to pay for their risk they
would say we don't want them any more, climate change, and we might actually have some action.
Munther (35:16):
This is fascinating I would say, because I will tell you that the
impression you would get from people is that insurance companies are making a killing.
Kerry (35:25):
No they aren’t, absolutely not they're - the only sector of the industry that's making
money and not that much is reinsurance, and that's because they're global and they're
not subject to these US regulations, but when they raise their prices that means insurance
companies have to pay more, then they have to go to the regulators and the regulators don't
(35:48):
want them to go out of business, but look at all the companies who have pulled out of Florida and
California because they simply can't afford to stay there. That's another consequence of this.
Kelvin (36:00):
When we look at the kind of information that insurance and reinsurance companies get and
use, a lot of times they go out and they contract with risk providers, risk information providers I
should say who do run catastrophe models they run all kinds of actuarial models and things as do the
insurance companies themselves, and when you talk to them what they will tell you is - ‘we're coming
(36:22):
to a fork in the road where we can't just keep raising rates’ - to Kerry’s point. They don't have
the freedom to do that, number one, but in some cases they can raise rates but they say - ‘there's
a cliff coming we don't know what to do about it.’ And so in talking with them about ‘what do
you need, what kind of information do you need?’ that's what we're really learning a lot about,
and what challenges they face, what the future should look like, because reinsurance contracts
(36:45):
are written only year by year. So they look down the the road maybe about a year or so,
we say ‘well should you look 10 years?’ well that's that's kind of a long time horizon. So
their thinking is changing in terms of their time horizon and how they actually make decisions,
but at least the ones I've spoken to which are quite large, both reinsurance and insurance,
they're like - we just can't keep doing this, there's got to be another model,
(37:05):
and Kerry I'm sure knows a whole lot more about that than I do,
being on the boards, but they are searching for something and they haven't found it yet.
Munther (37:13):
So I think that this is fascinating because, back to the signaling issue,
because I mean, I think, well, price is a very effective way of signaling,
maybe even direct signaling will be helpful, I don't know, some people may take it if you
just inform them them that in fact this is what's going on in your area that you're
becoming a higher and higher risk. At least some people, I would say a little bit over
(37:35):
50 percent of the American people have fear from climate change and the effect it has on
some hazards. T hey don't know how to pin it down and there's a debate and they feel that
they're on one side of that debate but there's also the climate change deniers who just don't
think any of that is true and so forth and they would use different arguments. Are we doing this
(37:57):
debate enough justice in terms of communicating to the public what they need to know? Because I
think you cannot communicate the models that you develop, because those too complicated, but what
do you communicate for them to be convinced that there is an issue and the level of uncertainty
around that particular issue and then preparing them to take some responsibility in the outcome.
Kerry (38:20):
The irony there is a lot of the sort of climate denial community,
if you want to call them that, are politically conservative and are not in favor of too much
government regulation. Fine, so you say let's deregulate insurance, and then places people's
insurance rates in many places would go through the roof, in other places they would go down.
(38:45):
I mentioned Kansas, they would go down in Kansas because those people don't realize
they're paying too much and subsidizing the folks in Palm Beach. So they may not resist
the notion that we should communicate risk among other things through pricing,
pricing works better than prostelytizing, I've discovered, but I think people have known that
(39:08):
for a long time. The problem is that if you just went out there today and said ‘we're not
going to regulate insurance premiums’ it would be wholesale chaos, because people would lose
their homes right away, they just would not be able to afford to play pay the insurance because
the insurance is so high their property value would go down okay, and this is a big problem
(39:32):
to affect that transition in my profession or in the insurance world this is sometimes referred to
as ‘managed retreat.’ How do you get from where we're at to where we ought to be in this problem,
and there have been ideas kicked around - it's complicated. One thing you could do is have for
sufficiently poor people, and they may not be that poor, have governments for a while subsidize their
(39:57):
insurance rates but have that subsidy decline over time to give them sort of a incentive and
a period of time to literally retreat from the dangerous places. I like to say it's one of the
few areas I know about where free market gurus and coastal ecologists make natural bedfellows.
(40:19):
*laughter*Kerry: ‘Free up the coastline’.
Liberty (40:23):
You know, it's a great thought that we'd change all of this in America
and people would sacrifice and, I don't know, in some ways the economy takes a huge hit - but
like does it really matter that much if we do it in the US and everybody else doesn't do it?
Because I don't see China and Russia and India - I mean, no one else is changing, so are we actually
(40:45):
going to be able to make enough of a dent that it even matters? Now I know it's not, you know,
two wrongs make a - you don't want to just do something just because everybody else is doing it,
but is it actually going to do any good unless the whole - as you said Kerry,
unless everybody comes together and says ‘we need to do something.’
Kerry (41:03):
Well, I think what you'd have to bet on is that after climate damages start to really
mount up, there's a strong incentive all around the world to manage this,
right, to deal with it. I’m putting aside the mitigation problems. We’d all like to do that,
but if we're not doing that these climate damages will just continue to soar,
(41:25):
and so it's in most countries’ economic interest to do something about that,
especially if they have - as is true not just in the US but all over the world - people migrating
from less risky to more risky places. So if we started to implement policies to reverse that
demographic trend or at least stop it, I think it would be imitated by other countries if it works,
(41:49):
right? So it's in their economic interests, If it's not, they probably won't do it.
Kelvin (41:54):
And I think as things become more obvious to people, what Kerry said will become
increasingly true. Part of the challenge you have is that some people who get out there and
have very loud voices, they're saying things that that just are kind of ridiculous. So there may be
famous people or whatever but they're they're not speaking the truth, and yet on the other hand if
you read the literature the climate modeling community is extremely open and transparent
(42:17):
about all of what's wrong with climate models, I mean they’ve got huge biases they've got real
challenges, but yet that doesn't mean they're not useful. So there's a difference between, sort of
to your area Liberty, sort of statistical accuracy and things like that and practical
value. And so we still can get value out of that, and I think as a scientific community
we're pretty honest about what we know and what we don't know and what our limitations are. It’s
(42:40):
hard for some people to think beyond - well this is 30- 40 years down the road, but as Kerry said,
as this stuff becomes more economically obvious, that China which emits you know roughly twice the
greenhouse gases that we do will hopefully kind of fall in line. The challenge there though is,
of course, with the Belt and Road Initiative and things like that they're building coal-fired
power plants in Southeast Asia that don't get accounted to China in its emissions
(43:04):
indexes. So they're actually well over what what we would think as far as the nation,
but they've got the attachment to all these different countries and things like that
which geopolitically is a real challenge. So these are interesting challenges but like
Kerry said it's a global challenge, we've got to think globally in terms of how we do this,
but at the same time when I was at the White House, what I would say to people was we can't
(43:26):
wreck our own economy in the process. We do that then then a lot of other stuff that is important
in terms of the vitality of America and its leadership role sort of goes down with it,
and we can't afford to do that. So we have to take this really thoughtful approach and and
the kinds of things Kerry’s talking about with the insurance industry and stuff, there are some some
actions that could be taken policy-wise that that might really help get us on a good pathway there.
(43:57):
[Music]
Liberty (43:57):
So Munther, what what was your sort of takeaway from this because I feel like
we were with two titans of this - of the climate change weather prediction world.
Munther (44:06):
Well, it was really interesting. I would say there's one part that I am familiar with,
and so - and them talking about the modeling and the and the limitations of accuracy and all of
that obviously I've heard all of that before so it was interesting, but the conversation
about insurance was actually surprising to me and it's not something that I have thought
(44:27):
about or known enough about. As I said, I think a lot of people have the perception
that insurance is making a killing and the story about how deregulation and being against
climate change you know come together in an interesting way right, and so I thought that
was interesting and it's kind of like brought some ideas for me to think about a little bit
Liberty (44:48):
I know every time I pay my insurance bill I'm like ‘god, I’m in the wrong business’
and it - as Kerry said, prostelytizing doesn't work as well as money in your pocket, it makes so
much sense like, I gotta say I know a fair amount of people who think the climate change stuff is
nuts but if all of a sudden their insurance bills increase six times they'd be all for stopping
(45:11):
climate change. So it's a very interesting concept that I just honestly never thought about.
Munther (45:16):
Yeah, and I think that it sort of brings up trying to understand from economists so to
speak how one can balance this. It doesn't always have to be as part of the premium
right and so is the signaling done in a different way? So I really need to understand this kind of
interesting aspect of signaling that he was talking about, but it is fascinating also to
(45:39):
try to understand how much is going on and the sort of conversations that are happening between
governments and private sector and academia and so forth, it’s a really interesting topic.
Liberty (45:53):
I think my final thought on it is that - what Kelvin said about how the actual,
the climate community that's studying this is actually very honest about the uncertainty
and where the issues in the models and stuff, but then you have the problem of
people fear-mongering and scaring people which actually does the opposite and makes people not
(46:15):
pay attention to it anymore and think it's all nonsense. So that communication aspect
I think is is just so important and it's such an interesting thing to highlight.
Munther (46:24):
Yeah, I totally agree. I mean, I think that even the people that are prone
to just disagree with climate change will use the uncertainty and the conversations
that the communities have as evidence that they don't know what they're talking about,
but actually it's evidence that this is science done well. So that's the signaling
(46:46):
and the argument with the communities and it's fascinating that people are getting
more and more sophisticated because of that, but still we have an issue.
[Music]
Thank you for listening to this month's episode of Data Nation from the MIT Institute for Data,
(47:08):
Systems, and Society. You can learn more about IDSS and listen to previous episodes at our
website idss.mit.edu or wherever you get your podcasts. Thank you to our producers Tina Tobey
Mack and assistant producers Ben Stull and Maria Brooks. Don't forget to leave us a review and
follow us on X at MIT IDSS to stay informed. Thank you again for listening to MIT’s Data Nation.
[music]
Today on Data Nation, we're speaking with Kerry Emanuel, co-founder of the Lorenz Center at MIT,
and Kelvin Droegemeier, former director of the Office of Science and Technology Policy,
a position otherwise known as Chief Science Advisor to the President of the United States,
in order to dive deeper into natural disasters, climate change, and extreme weather prediction
(00:29):
technology. I’m Liberty Vittert, professor at Washington University in St. Louis, and my co-host
is Munther Dahleh, William A. Coolidge professor in electrical engineering and computer science at
MIT. In 2024, Americans all over the country faced loss and damage from natural disasters
and extreme weather events, from LA wildfires to hurricanes Helene and Milton in the Southeast.
(00:54):
The NOAA’s National Center for Environmental Information estimated that extreme weather
events in America caused almost $200 billion worth of damages in 2024 alone. One year prior,
the US Census estimated that 2.5 million Americans were displaced from their homes
by weather-related disasters. With millions of Americans at risk of losing their homes
(01:19):
and communities due to weather conditions, climate experts and scientists have been
working relentlessly to develop premeditated and predictive weather technologies to limit the
damage from natural disasters. With 2024 recorded as the warmest year on record according to NASA,
climate change and natural disasters will continue to be a problem for everyone on Earth. With the
(01:44):
challenges of global warming and the pressures to prevent widespread damage to Americans. We
are speaking with Kelvin Droegemeier and Kerry Emanuel to better understand what legislative
and scientific changes can be made to combat future issues from natural disasters in America.
Munther (02:03):
You know, I think the first question that came to my mind is about the fires in LA,
150,000 people displaced. Did we mispredict what happened that got us to that point and
is this about sort of advancing the knowledge and the science behind the
modeling? What's missing and put us in the right track on this particular issue.
Kerry (02:24):
I can say that there's a whole kind of cottage industry in
what's called extreme event attribution, to what extent can you take anything like
a big hurricane or a wildfire of the kind that affected Los Angeles and say anything
scientifically credible about the extent to which climate change has had an effect on that,
(02:48):
and I don't know what exactly the answer is for that particular event. Although climate
models generally do predict that conditions for wildfires in the West writ large will becoming
more favorable for those fires, but that's not to say that particular fire. I want to say that
(03:09):
that when I look at problems like that I look at it in the context of a non-climate trend,
which is a strong demographic trend. People all over the United States to be moving from
less risky to more risky areas. It’s a trend that arguably at least partially takes place
because we inadvertently subsidize that trend, and that's something that you might say that
(03:35):
climate change is a threat multiplier of that, as people move into riskier places.
That's something that we could do something about from a policy perspective, and probably
should irrespective of whether climate change is becoming more dangerous, but certainly it is.
Kelvin (03:53):
I’d like to build on what Kerry said about attribution science. It is a science of
attributing a particular event or class of events to other forcing phenomena like climate change and
other things like that, and it's gotten quite good in in the past several years, it's advanced rather
rapidly and now there's a technique called pseudo global warming where when we run our
(04:13):
climate models Earth system models for very long periods of time we start them with sort of a some
reference climate state and we run them to kind of an equilibrium and then we start turning on
various forces and things like that, but to your point about the the wildfires on a particular day
or particular part of the year how do we do that? So one of the ways to do that is take a given
(04:34):
event that has happened and then look at what the climate would have been like prior to that event
and then take that delta and resimulate that event and really not be able to say just a class of
events can be attributed to some other phenomena like climate change but a specific event. Now,
as Kerry said you it's very difficult to do that in a lot of cases. Wildfires, of course,
(04:55):
have multiple dimensions to them and it's very much like my own experience in working a lot of
aircraft accidents, commercial accidents, it's not typically one causal factor it's typically
multiple things line up and they line up a particular way to cause something to happen. So it
is a difficult thing to tease out, but I will say that I think we are getting better at doing that
and so I think from a policy perspective the real key is to understand what things we really can
(05:19):
influence on the short term and the long term, and then take actions in a policy sense and other ways
to look at effectuating some type of of change to prevent those things from happening in the future.
Munther (05:29):
So, in general, predicting these extreme events is very difficult, and then there's the
question of having the right emergency response system in place. It’s not so much about having the
accuracy, and for example when we think about the failure of the outages of the power grid,
again those are the kind of things where it's very difficult to predict the cascaded failures
(05:52):
that can occur over there, but there are some simple principles - and we don't do that very
well - but there are simple abstracted models that can give you an idea of when
to start mitigating some of that risk. Is this analogy extendable to situations like
fires or hurricanes or - in the extreme case, not in the more predictable cases?
Kelvin (06:16):
Well, I think pre-positioning assets and anticipating an event is extremely important,
and one of the things that that is being worked on now around the world is to take
our Earth system models which operate relatively coarse meshes in terms of the fidelity that they
have to resolve fine scale features and bring them down essentially to the scale
(06:37):
of what we use now in numerical weather forecasting, over say a two or three week
period. When you're running an Earth system model for multiple years or multiple decades,
we don't have the computational power to run at that fine fidelity which provides the much more
localized decision-making information that people would need to anticipate events and prepare for
them. So it's a difference between very strategic positioning and very tactical positioning. So one
(07:02):
of the the things that is is hoped for for the future, and of course artificial intelligence
influencing this as well, is to really be able to provide that very localized fine-scale
information over multiple decades actually for things like agriculture, for transportation,
for the kinds of things we're talking about here. We’re not there yet, and we've used the concept
(07:22):
of downscaling where you actually provide a much finer mesh of the model over some area,
and that is really excellent thing to do but there are some limitations with that. So there are about
over a dozen groups around the world now that are working on these what are called kilometer-scale
Earth system models, and moving from what we have been using historically to that type of
fine-scale is a tremendous challenge, it takes a long time to do that. Plus, we actually then have
(07:48):
to figure out - how do we initialize those kinds of models, perhaps even using real observations,
which is really not something that is done now. So it's a very exciting area of science that is,
I think, helping position us to answer exact the kind of question that you posed. If you
talk to decision makers, you talk to people in the reinsurance industry, the insurance industry,
people who make these more localized decisions are - they will tell you ‘we're not really
(08:09):
getting the information that we need’ and so I think listening to those stakeholders and having
them drive the agenda to where we make sure that we're paying attention to what their needs are,
as well as understanding the science of the earth system which is extremely important.
We have to do both and they're very, very much congruent with one another.
Liberty (08:26):
Just to sort of follow on, Kerry, from something you mentioned about people moving to
sort of high-risk areas. I don't remember where I read it but you know it was like,
Miami's built on a swamp, there's going to be more damage when it says there's ten times the amount
of money, more damage. it's like well that's that's how it's going to work when you have a
(08:46):
lot more people living there, and there was that recent study that found that human-caused climate
change made Atlantic hurricanes, I think it was 18 miles per hour stronger in the last six years
and you had had said that the study makes and is similar to your predictions from almost 40 years
(09:06):
ago. So could you give us sort of a little bit more in-depth discussion on how climate change
is worsening natural disasters, or is it both, is it that it's worsening things but we also
have people living in Miami on a swamp, and what are what are your sort of future predictions for
the strength of natural disasters and is there hope, really for dampening effects in the future?
Kerry (09:31):
Well it's a long and interesting string of questions.
Liberty (09:34):
Sorry, it was my stream of consciousness here.
Kerry (09:37):
I think we have to take this hazard by hazard. I've worked a lot, as you pointed out,
on hurricanes, and the community gets together, the community of people who research hurricanes
and climate get together periodically and compare notes. I think there's pretty strong confidence
that hurricanes are becoming more intense. We’re beginning to see it in the satellite data.
(09:59):
They not necessarily becoming more frequent except in the Atlantic region and that is for
reasons that may not have a lot to do with global warming, and very importantly they rain more, and
what many people don't realize is that the major source of loss of life in hurricanes is not wind,
it's water, either from torrential rain-induced flooding or I should say and/or from storm surges.
(10:27):
So they're getting worse but that's compounding the other phenomenon you mentioned, and this could
be said of some other hazards as well, of people moving into places not not just living there,
people have lived in Miami for a while, but actually actively moving from less risky and
more risky places and this is very much on my mind because it's very hard to reverse that. Why
(10:51):
is that happening? Well there are all kinds of ways we subsidize people to live in places they
shouldn’t, and that is to protect them from the cost of that. It’s complicated and it has to do
with the way insurance is set up and regulated, and it sounds humane on short time scales but
it's catastrophic on long time scales. So, for example, all of the water damage which
(11:15):
is usually a lot of it from natural hazards like floods and hurricanes is insured by the federal
government through the national flood insurance program, that's been true since 1968, and because
it is a government enterprise it's captive to politics and people usually well-connected,
(11:37):
relatively well-to-do people living in dangerous places agitate for their rates to be kept low, and
indeed they are. So, for example, where I live on the coast of Maine it's not exactly Palm Beach but
it's wealthier than inland, we get in my county four times as much money payouts from the National
(11:59):
flood insurance program as we collectively put into it through premiums. In Western Kansas,
that ratio is reversed. People are putting four times as much premiums in as they're getting back,
and that's unfair I would say in the first place but worse than that it's encouraging
this migration. There are all kinds of other ways this happens. But on the good side of it,
(12:23):
and to add on to something that Kelvin said we actually have gotten collectively much much better
at warning people in the face of an immediate disaster, at least for some kinds of hazards like
hurricanes and floods, and the death toll from these things has gone way down in spite of the
fact or at the same time the damage is going way up, and we can do better and we should do better
(12:48):
and I don't think we're anywhere near that level of competence for wildfires but we're headed that
way. So it seems like a tragedy that we're able to save lives but also we're suffering a lot more
damage at the same time, that's the general view of the natural hazard landscape from where I sit.
Kelvin (13:10):
If I could just build on something that Kerry said and take it even to a
broader perspective, this is something that the research community an the atmospheric sciences and
meteorology and the climate sciences community began to really recognize probably twenty or
so years ago and that is that all the physical science and technology in the world isn't going
to prevent people from dying, isn’t going to prevent some of the catastrophic loss, and that
(13:32):
really we're dealing with the coupled human and natural systems. So the social-behavioral
sciences and the economic science dimensions of what we're talking about here, in particular
hurricanes, is extremely important. So now we're studying these problems really in their totality,
bringing in policy dimensions as well from the very beginning rather than just be at the end
of where it's like ‘okay we now understand the physical science of this phenomena,
(13:54):
now what do we do with that understanding?’ We have those folks at the table from the very
beginning, because this truly is a coupled, very entangled and interactive kind of system and so
to achieve the things that Kerry talked about is really, it means to understand human behavior, how
we communicate risk, how humans respond to risk and decision-making under stress, what evacuation
(14:14):
pathways exist and so on and so forth, and for example there's a high risk or some type of a
risk or a threat of say severe thunderstorms we do a great disservice by using that term because
risk depends on your own particular situation - where you are, where you live, is English
your first language, your risk is determined by a lot of factors compared to what the atmosphere is
(14:35):
going to do and so this is an area of scholarly work that is really getting a lot of attention
now to convey risk in a way that is sort of customized to your own particular situation at
the moment or moments preceding the event actually happening, which I think is going to add to what
Kerry said in terms of increasing the ability to prevent the loss of life and hopefully protecting
property and reducing loss. So I just thought it would be important to kind of mention that.
Munther (15:00):
I think that's really interesting, and at some level it's comforting to know
that less people are dying even though the damage in terms of cost is very high,
but the point is tailoring the knowledge of the situation to the particular person
I think it's really a fascinating idea. So to segue a little bit, people are scared,
(15:22):
a lot of people have these conversations about ‘what if the unexpected happens in my region?’ you
know it's like maybe earthquakes in places that don’t have earthquakes or a sudden flooding in a
different area and so forth. So we have that one aspect, there's all the science with the
science community and what they know and there's a mistrust, of course, between the communities that
(15:45):
we talk about and then there's the government and Kelvin you have been the head of the White
House Office of Science and Technology Policy and so there's the aspect of that which is
regulated and mandated by the government. How is that working out all together? All these
different components and what is the government actually doing to help in these situations?
Kelvin (16:05):
Right, no that's a really great question and ultimately the federal government
has responsibility to protect its citizens, to ensure economic vitality of the nation
and things like that. What a lot of times folks don't understand is that when you develop policy,
my role at OSTP was to make sure it was at the table, but policy actually is not just
(16:25):
what science says is the case, and of course we learned and saw in the pandemic the public
really got I think a peek behind the door to see how science really works, it's not here's
an immutable answer to this particular thing, do I wear a mask or don't I? This study said one thing,
this study contradicted and that's that's how science has always worked, but I don't think the
public generally understood that, they just wanted to know do I or don't I wear a mask, just tell me,
(16:47):
and so when you think about the policies that get formulated especially in situations like that,
science is at the table but there's so many other factors that come into play. There's economics,
there's certainly politics there's national security, all these kinds of things and you
see the policy come out the other side and you're like ‘well wait a second, that seems incongruent
with what we think we understand about about the natural world, about hurricanes or whatever’ so
(17:09):
it's a very complicated process and sometimes it doesn't turn out the way you would perhaps
hope it is because one part of that policy forcing dominates another, and so really all that we can
do as a scholarly research community is make sure that that we convey the facts as we best
understand them, we do it in such a way that that other folks can really take that up and use it and
(17:30):
then translate it into meaningful policy that is actionable, that has intended outcomes, and
doesn't have any serious unintended consequences which sometimes is is quite difficult, but I think
that the government works fairly well most of the time, but when you really start to understand how
policy happens and how the sausage is made then it begins to make sense of why some of
the things are the way they are and they're not the way you would think they should be.
Liberty (17:52):
Thinking about the future of all of this, Kerry, I saw a research article that you did sort
of proposing TC-GEN and the article described from my understanding TC-GEN is a machine learning
based framework that basically uses historical data sets and climate models to reproduce these
(18:14):
really high-resolution weather conditions, and I wanted to know in terms of the concepts of the
future, can this past data, especially given as you said with high-risk areas and how things are
changing so quickly and people are moving to these places, can it really help sort of reproduce these
conditions now, and do you see this technology as really helping disaster preparedness for regions?
Kerry (18:40):
In spite of the fact I was a co-author on that paper, I actually prefer the kind of
approach that Kelvin was talking about earlier, it's called downscaling. Let me try to explain
why I do that . So one of the problems we have is if you look at places that have 100 years of
good records of weather and also of damage that weather causes, you discover that over a very
(19:08):
long period of time if you accumulate damage from that hazard most of the damage comes from events
whose frequency is roughly once in a 100 years. The very frequent events we’re well adapted to
so they don't cause much damage and if they're too rare, well they're so rare that they don't
contribute in the long run. Well 100 years, to get a really robust estimate of how much damage a 100
(19:34):
year event would occur you need about a thousand years of records. Well, we don't have that,
in fact we're lucky if we have a hundred. So even if the climate weren't changing, we have a problem
that we're not able through the historical record to really see many of the events that in the long
run, like the Los Angeles fires, the rare events that cause a lot of damage, and so I've been on a
(19:58):
campaign to try to use physics through models to evaluate today's risk. That whole problem is
made much worse by climate change and I think it's true to say for many phenomenon that even
if we had a thousand years of beautiful records, arguably they're not relevant to today because of
(20:19):
the climate change that's already happened, and insurance companies are very rapidly waking up
to this fact. All of their risk models, all the decisions they make and a lot of cities and towns
and states are making are based on historical data which is no longer relevant to today. So this
(20:39):
is part of the reason that we're seeing so much damage is that we're going outside this envelope,
but as great as artificial intelligence is it usually fails when you try to push it outside the
set of data that you use to train it, and that's automatically true with climate change. You just
(21:00):
don't have data pertinent to today's climate, you don't have nearly enough of it to do that.
So I think downscaling is a way out of that and I've tried very hard to do that in the case of
of hurricanes, and we are seeing that in many places including the Carolinas, the risk today
is much higher than it was 20 or 30 years ago and yet it's not being reflected in insurance rates,
(21:25):
it's not being reflected in construction practices and building codes. So that's where
a lot of this damage is coming from, we're just not adapting fast enough to the changing climate.
Munther (21:38):
But if I jump in and say that - trying to understand what you meant by bringing in physics,
because in the world of Engineering in general what you said is absolutely true
that is basing models on pure historical data can actually miss the future. But I
think one important aspect of this is modeling all the components and
(21:59):
all the pieces and understanding the the relationships between them right,
and so the question is if temperature and heat and other factors also affect the global phenomenon,
why aren't the models developed to take into consideration all these inputs and then simulate
and learn potentially the underlying physics which has like been demonstrated in other fields.
Kerry (22:24):
Well, there's a lot packed into your very interesting question. So one of the problems is
the word model. It means very different things to different people, even to different sets of
scientists. So a numerical weather prediction model of the kind that Kelvin talked about a
few minutes ago is a very, very different piece of machinery than an economic model. That is,
(22:49):
in the first case we actually do know the laws of physics, but we can't afford to integrate them at
high enough resolution to capture even phenomena like hurricanes, and we have to get clever and
embed fine-scale models within these - there are ways of doing that. But the enormous advantage
(23:10):
if you can do that is the laws of physics are invariant, they don't change because the
climate's changing. So climate models are models that we would like to see resolve phenomena that
are on scales of several hundred kilometers or larger, they might be able to handle heatwaves. In
practice, because their resolution is limited, we have to do something called parameterization for
(23:38):
things like small-scale phenomena like convection and different models do it differently, and so one
of the things we see when we downscale different climate models is we get very different answers.
So it's not the laws of physics that are to blame for that, it's the way we encode them in models.
Kelvin (23:58):
And the one advantage that we hope to gain when we take these models that Kerry’s
talking about, and instead of having to downscale over a particular area, we actually have a uniform
resolution or nearly so across the globe, is that we then get the feedbacks across the scales. Kerry
talked about convection, these major convective systems that then feed back to the large scale,
(24:18):
the large scale influences the direction of the convective system, that cross-scale interaction
both in time and space is very important and it's not something that we really capture even
with the downscaling models, as phenomenal as they are. I've seen downscale simulations that
you’d swear you're looking at a satellite loop, it's it's just phenomenal. So the good news is
the science is really advancing, we're learning more but Kerry’s really made an important point
(24:41):
here that AI is not a singular solution. Now in weather prediction, in weather forecasting we're
seeing dramatic effects there because we do have a very, very long record and multiple records of
predictions and things like that so there's a lot more information out there to train models with,
and the interesting thing when you think about machine learning, when you train a
model it sort of knows the future. When we're using a physics-based model we initialize it
(25:04):
with what's happening up until the current time that we start the model and then it has
to forecast the future. The AI has the advantage of knowing what the future is because it's been
trained on a lot of different futures but it has the inherent limitations, as Kerry said,
that if you try to push it too far beyond the training data set you run into problems. So
there'll be a lot of folks employed in this arena for a long time because there's a lot of work to
(25:26):
be done, but a lot of it is quite promising, but there are no silver bullets, that's what Kerry
and I found in our combined 85 or 90 years of of work, there is really no such thing.
Munther (25:36):
This is really interesting I'm just wondering I'm going to push the point of data
science and machine learning on this a little bit, because what we found for example in the last five
years our ability - take the gene folding problem, okay, the gene folding problem, I
mean you're talking about a genome of over 20,000 genes. Even if each one could take a binary value,
(26:02):
the possibilities is two to the 20,000 which is enormous and the gene folding problem is
mapping that gene structure to a fold that then maps a function and what we've been able to do,
I think, with a lot of the machine learning and clever - and I would say and that was
used before - clever ideas because it's not just a brute force application is to create 30 million
(26:27):
possible new data sets just by simulation, and when we were at only 150,000 in the last 30
years with with experimentation, and with a lot of confidence that these simulated data are actually
quite accurate. So is the community embracing this kind of thinking into these geochemical models
(26:50):
and complex sort of high-dimensional models? Is that what you're seeing in your community or not?
Kelvin (26:56):
I can tell you here at the University of Illinois we have a whole major center,
a biofabrication center that does exactly that. It looks at - instead of doing Monte Carlo,
you're doing thousands and thousands of realizations - you use AI to really bring that
number down and then AI determines what the next iteration ought to be. So it's a circular kind
of a thing where you're really informing the next calculation based upon the results that you have,
(27:20):
and I'm certainly not a biologist but I've talked to these folks in molecular and cellular
biology and they're doing really extraordinary things. It’s very much more rapidly allowing us
to understand how - your protein folding and genetic structure and things like that - to
actually fabricate small molecules and that sort of thing to cure diseases,
(27:41):
to deal with inflammation and that sort of thing. So it is certainly happening to
to my degree of awareness in that area right here on my own campus just down the street.
Kerry (27:49):
Let me add briefly to that, it seems like every other article in today's journals in
my profession, in our profession is about AI and machine learning, so that's maybe not the best -
Liberty (28:02):
I feel like that’s all professions at this point.
Kerry (28:03):
It’s all professions, but I would also say that it's very good for some things like
making many, many, many replicates of weather in today's climate and as I mentioned a while ago,
just quantifying natural hazard risk in today's climate is a huge step forward,
before we actually start worrying about the future. The fact that today is very different
(28:26):
from fifty years ago is is really, really important. The other thing I would say is
kind of a no-brainer for AI is that these big models, numerical weather prediction models,
climate models, because of their need to parameterize small-scale processes,
partly because of that, they have lots and lots of knobs to turn, and so if you - that
(28:48):
you can tune physically within some plausible range - to do that manually is just impossible,
there far too many combinations like your genome example, and so AI can help us optimize those
models. I personally think that some combination of deterministic models, ensembles of such models
(29:12):
enhanced by AI will probably be the way we make the most rapid progress, but I'm just guessing.
Kelvin (29:20):
I think Kerry is spot on with that those comments, and to the last point, when you're
looking at these very sophisticated Ferrari-level Earth system models that are just extraordinary,
we're driving them like a 1950’s small car because we're constrained by the computational resources
we have. It turns out that if you really turn those things loose on the Autobahn,
(29:43):
about 95 percent of the computational resources would be taken up by what we call the dynamics
of the model, the fluid physics that Kerry talked about. Roughly five percent with these
physical parameterizations of precipitation and clouds and so on, but yet to Kerry’s earlier
point those things are extraordinarily important and that's where AI, although it doesn't really
(30:04):
improve the efficiency of the model that much, it really improves as he just said the accuracy and
ability to capture these very complicated but very important processes in a much more accurate way,
that plus producing ensembles very, very inexpensively. But as far as you're speeding
up the core component of the model not so much. GPU’s do a very good job of that but really AI
(30:28):
is is really geared toward the kinds of things that Kerry talked about.
Liberty (30:31):
So I have to ask something that I feel like we've touched on this whole way through. So,
my parents have a house in Fort Myers, well they had a house in Fort Meyers Beach,
and the hurricane two years ago, Ian, wiped it out, and so they're thinking about rebuilding
(30:52):
but trying to get insurance is almost impossible there, and Kelvin, thinking about what you said
about individual risk and the differences, for them it was fine. They were able to get out
they had another home to go to, they were fine. Some of their neighbors though, they were either
really sick so it was hard for them to get out, or they had pets that they didn't want to leave
(31:15):
and the shelters wouldn't take. That concept of individual risk is very, very interesting, but in
the same sense of how hard it is to get insurance in some places, and then Kerry as you were saying
how in other places the insurance companies haven't kept up, do the insurance companies
know something we don’t? Are they predicting - are they a lot better at predicting these
(31:38):
natural disasters coming forward than we are? I mean, because it's so hard to get it, are they
are they just doing that based upon past data or are they predicting new things that we don't know?
Kerry (31:50):
No. I can tell you with some authority because I sit on the boards of two insurance
companies and half for twenty years, it's a much different situation than most policy holders,
because - policy holders don't like insurance companies because they're being nickeled and
dimed and there are these exclusions and - it's all true, but the enemy isn't necessarily the
(32:14):
insurance company, it’s the way it's regulated. And insurance has always been regulated for the
same reason in principle that we regulate banks. We have to ensure they're solvent so when the big
storm comes they don't pack up and say ‘we're bankrupt’ and leave you high and dry. In fact,
something has been happening for a long time with insurance that goes exactly counter to that goal,
(32:39):
that is regulators started to regulate premiums. Now this sounds very esoteric,
and it doesn't sound like I'm answering your question, but I'm going to. And that has become
captive to a lot of of political pressure in every state I'm aware of, and so what what's happened
is that insurance companies don't have the degree of freedom - if they had a better model that told
(33:04):
them the risk was worse than they think it is, they can't just go ahead and adjust the premiums,
which would have the beneficial effect of sending a message to homeowners that this place is getting
riskier and maybe encouraging them to move or or take some measures to make their dwelling a
(33:24):
little bit more resilient and so forth. But we're not letting that message go through,
and so the the result is that the profit margin of insurance companies is razor thin,
razor thin. It’s a terrible business to be in and it's not being allowed to do the job it should be
doing of communicating risk through pricing. So we have this really, really bad situation,
(33:47):
and one of the problems is - you'll notice this is happening in Los Angeles it happened in Fort
Meyers, Liberty where your parents lived, it happens everywhere after disaster - mostly just
go right back and rebuild the same thing they had before, that's the instinct to do that, If it were
a rational insurance market, they couldn't do that because they couldn't afford to insure it if they
(34:11):
did. So if we did that one thing by gracefully deregulating insurance - and I'm not a free
market ideologue but in this one instance - people couldn't afford to rebuild their property in a way
that puts that property at risk. The message would come through you do that you're going to
(34:32):
have to pay through the nose for insurance, but if you do it this way, if you make your house
fire-resistant in this way and that way, you'll have a much lower premium, we're not letting that
happen. To me that is the heart of the problem, in fact it's the heart of the whole response to
climate change problem. The reason that people aren't out in the streets agitating for our
(34:53):
government and world governments to do something about climate change is we're insulating them
from the risk. We’re not letting them feel it. Your house got blown down, okay we replace it no
strings attached. This isn't working for society and if people had to pay for their risk they
would say we don't want them any more, climate change, and we might actually have some action.
Munther (35:16):
This is fascinating I would say, because I will tell you that the
impression you would get from people is that insurance companies are making a killing.
Kerry (35:25):
No they aren’t, absolutely not they're - the only sector of the industry that's making
money and not that much is reinsurance, and that's because they're global and they're
not subject to these US regulations, but when they raise their prices that means insurance
companies have to pay more, then they have to go to the regulators and the regulators don't
(35:48):
want them to go out of business, but look at all the companies who have pulled out of Florida and
California because they simply can't afford to stay there. That's another consequence of this.
Kelvin (36:00):
When we look at the kind of information that insurance and reinsurance companies get and
use, a lot of times they go out and they contract with risk providers, risk information providers I
should say who do run catastrophe models they run all kinds of actuarial models and things as do the
insurance companies themselves, and when you talk to them what they will tell you is - ‘we're coming
(36:22):
to a fork in the road where we can't just keep raising rates’ - to Kerry’s point. They don't have
the freedom to do that, number one, but in some cases they can raise rates but they say - ‘there's
a cliff coming we don't know what to do about it.’ And so in talking with them about ‘what do
you need, what kind of information do you need?’ that's what we're really learning a lot about,
and what challenges they face, what the future should look like, because reinsurance contracts
(36:45):
are written only year by year. So they look down the the road maybe about a year or so,
we say ‘well should you look 10 years?’ well that's that's kind of a long time horizon. So
their thinking is changing in terms of their time horizon and how they actually make decisions,
but at least the ones I've spoken to which are quite large, both reinsurance and insurance,
they're like - we just can't keep doing this, there's got to be another model,
(37:05):
and Kerry I'm sure knows a whole lot more about that than I do,
being on the boards, but they are searching for something and they haven't found it yet.
Munther (37:13):
So I think that this is fascinating because, back to the signaling issue,
because I mean, I think, well, price is a very effective way of signaling,
maybe even direct signaling will be helpful, I don't know, some people may take it if you
just inform them them that in fact this is what's going on in your area that you're
becoming a higher and higher risk. At least some people, I would say a little bit over
(37:35):
50 percent of the American people have fear from climate change and the effect it has on
some hazards. T hey don't know how to pin it down and there's a debate and they feel that
they're on one side of that debate but there's also the climate change deniers who just don't
think any of that is true and so forth and they would use different arguments. Are we doing this
(37:57):
debate enough justice in terms of communicating to the public what they need to know? Because I
think you cannot communicate the models that you develop, because those too complicated, but what
do you communicate for them to be convinced that there is an issue and the level of uncertainty
around that particular issue and then preparing them to take some responsibility in the outcome.
Kerry (38:20):
The irony there is a lot of the sort of climate denial community,
if you want to call them that, are politically conservative and are not in favor of too much
government regulation. Fine, so you say let's deregulate insurance, and then places people's
insurance rates in many places would go through the roof, in other places they would go down.
(38:45):
I mentioned Kansas, they would go down in Kansas because those people don't realize
they're paying too much and subsidizing the folks in Palm Beach. So they may not resist
the notion that we should communicate risk among other things through pricing,
pricing works better than prostelytizing, I've discovered, but I think people have known that
(39:08):
for a long time. The problem is that if you just went out there today and said ‘we're not
going to regulate insurance premiums’ it would be wholesale chaos, because people would lose
their homes right away, they just would not be able to afford to play pay the insurance because
the insurance is so high their property value would go down okay, and this is a big problem
(39:32):
to affect that transition in my profession or in the insurance world this is sometimes referred to
as ‘managed retreat.’ How do you get from where we're at to where we ought to be in this problem,
and there have been ideas kicked around - it's complicated. One thing you could do is have for
sufficiently poor people, and they may not be that poor, have governments for a while subsidize their
(39:57):
insurance rates but have that subsidy decline over time to give them sort of a incentive and
a period of time to literally retreat from the dangerous places. I like to say it's one of the
few areas I know about where free market gurus and coastal ecologists make natural bedfellows.
(40:19):
*laughter*Kerry: ‘Free up the coastline’.
Liberty (40:23):
You know, it's a great thought that we'd change all of this in America
and people would sacrifice and, I don't know, in some ways the economy takes a huge hit - but
like does it really matter that much if we do it in the US and everybody else doesn't do it?
Because I don't see China and Russia and India - I mean, no one else is changing, so are we actually
(40:45):
going to be able to make enough of a dent that it even matters? Now I know it's not, you know,
two wrongs make a - you don't want to just do something just because everybody else is doing it,
but is it actually going to do any good unless the whole - as you said Kerry,
unless everybody comes together and says ‘we need to do something.’
Kerry (41:03):
Well, I think what you'd have to bet on is that after climate damages start to really
mount up, there's a strong incentive all around the world to manage this,
right, to deal with it. I’m putting aside the mitigation problems. We’d all like to do that,
but if we're not doing that these climate damages will just continue to soar,
(41:25):
and so it's in most countries’ economic interest to do something about that,
especially if they have - as is true not just in the US but all over the world - people migrating
from less risky to more risky places. So if we started to implement policies to reverse that
demographic trend or at least stop it, I think it would be imitated by other countries if it works,
(41:49):
right? So it's in their economic interests, If it's not, they probably won't do it.
Kelvin (41:54):
And I think as things become more obvious to people, what Kerry said will become
increasingly true. Part of the challenge you have is that some people who get out there and
have very loud voices, they're saying things that that just are kind of ridiculous. So there may be
famous people or whatever but they're they're not speaking the truth, and yet on the other hand if
you read the literature the climate modeling community is extremely open and transparent
(42:17):
about all of what's wrong with climate models, I mean they’ve got huge biases they've got real
challenges, but yet that doesn't mean they're not useful. So there's a difference between, sort of
to your area Liberty, sort of statistical accuracy and things like that and practical
value. And so we still can get value out of that, and I think as a scientific community
we're pretty honest about what we know and what we don't know and what our limitations are. It’s
(42:40):
hard for some people to think beyond - well this is 30- 40 years down the road, but as Kerry said,
as this stuff becomes more economically obvious, that China which emits you know roughly twice the
greenhouse gases that we do will hopefully kind of fall in line. The challenge there though is,
of course, with the Belt and Road Initiative and things like that they're building coal-fired
power plants in Southeast Asia that don't get accounted to China in its emissions
(43:04):
indexes. So they're actually well over what what we would think as far as the nation,
but they've got the attachment to all these different countries and things like that
which geopolitically is a real challenge. So these are interesting challenges but like
Kerry said it's a global challenge, we've got to think globally in terms of how we do this,
but at the same time when I was at the White House, what I would say to people was we can't
(43:26):
wreck our own economy in the process. We do that then then a lot of other stuff that is important
in terms of the vitality of America and its leadership role sort of goes down with it,
and we can't afford to do that. So we have to take this really thoughtful approach and and
the kinds of things Kerry’s talking about with the insurance industry and stuff, there are some some
actions that could be taken policy-wise that that might really help get us on a good pathway there.
(43:57):
[Music]
Liberty (43:57):
So Munther, what what was your sort of takeaway from this because I feel like
we were with two titans of this - of the climate change weather prediction world.
Munther (44:06):
Well, it was really interesting. I would say there's one part that I am familiar with,
and so - and them talking about the modeling and the and the limitations of accuracy and all of
that obviously I've heard all of that before so it was interesting, but the conversation
about insurance was actually surprising to me and it's not something that I have thought
(44:27):
about or known enough about. As I said, I think a lot of people have the perception
that insurance is making a killing and the story about how deregulation and being against
climate change you know come together in an interesting way right, and so I thought that
was interesting and it's kind of like brought some ideas for me to think about a little bit
Liberty (44:48):
I know every time I pay my insurance bill I'm like ‘god, I’m in the wrong business’
and it - as Kerry said, prostelytizing doesn't work as well as money in your pocket, it makes so
much sense like, I gotta say I know a fair amount of people who think the climate change stuff is
nuts but if all of a sudden their insurance bills increase six times they'd be all for stopping
(45:11):
climate change. So it's a very interesting concept that I just honestly never thought about.
Munther (45:16):
Yeah, and I think that it sort of brings up trying to understand from economists so to
speak how one can balance this. It doesn't always have to be as part of the premium
right and so is the signaling done in a different way? So I really need to understand this kind of
interesting aspect of signaling that he was talking about, but it is fascinating also to
(45:39):
try to understand how much is going on and the sort of conversations that are happening between
governments and private sector and academia and so forth, it’s a really interesting topic.
Liberty (45:53):
I think my final thought on it is that - what Kelvin said about how the actual,
the climate community that's studying this is actually very honest about the uncertainty
and where the issues in the models and stuff, but then you have the problem of
people fear-mongering and scaring people which actually does the opposite and makes people not
(46:15):
pay attention to it anymore and think it's all nonsense. So that communication aspect
I think is is just so important and it's such an interesting thing to highlight.
Munther (46:24):
Yeah, I totally agree. I mean, I think that even the people that are prone
to just disagree with climate change will use the uncertainty and the conversations
that the communities have as evidence that they don't know what they're talking about,
but actually it's evidence that this is science done well. So that's the signaling
(46:46):
and the argument with the communities and it's fascinating that people are getting
more and more sophisticated because of that, but still we have an issue.
[Music]
Thank you for listening to this month's episode of Data Nation from the MIT Institute for Data,
(47:08):
Systems, and Society. You can learn more about IDSS and listen to previous episodes at our
website idss.mit.edu or wherever you get your podcasts. Thank you to our producers Tina Tobey
Mack and assistant producers Ben Stull and Maria Brooks. Don't forget to leave us a review and
follow us on X at MIT IDSS to stay informed. Thank you again for listening to MIT’s Data Nation.
Popular Podcasts
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!
The Breakfast Club
The World's Most Dangerous Morning Show, The Breakfast Club, With DJ Envy And Charlamagne Tha God!
The Joe Rogan Experience
The official podcast of comedian Joe Rogan.